mynn 0.1.2

Hobby no_std neural network.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35

#[cfg(not(feature = "f32"))]
use core::f64::consts::E;
#[cfg(not(feature = "f32"))]
use libm::pow;

#[cfg(feature = "f32")]
use core::f32::consts::E;
// Micromath works better on smaller 8 bit MCUs where we would be using 32 bits  
#[cfg(feature = "f32")]
use micromath::F32Ext; 

use super::Float;


/// Helper container type holding the closures for the activation function and the derivative. 
/// 
/// Used for forward and backwards propagation in the neural network. 
pub struct Activation<'a> {
    pub function: &'a dyn Fn(Float) -> Float,
    pub derivative: &'a dyn Fn(Float) -> Float
}

/// Sigmoid activation function, used a lot in the examples and tests. 
#[cfg(not(feature = "f32"))]
pub const SIGMOID: Activation = Activation {
    function: &|x| 1.0 / (1.0 + pow(E, -x)),
    derivative: &|x| x * (1.0 - x)
};

/// Sigmoid activation function, used a lot in the examples and tests. 
#[cfg(feature = "f32")]
pub const SIGMOID: Activation = Activation {
    function: &|x| 1.0 / (1.0 + E.powf(-x)),
    derivative: &|x| x * (1.0 - x)
};